PRIMECLUSTER and PRIMEQUEST Design Guide
Contents
1. quick recovery of operation For details see 3 2 1 Reserved SB in the Administration Manual Reserved SB 4 Partition 0 Partition 1 Partition 0 Partition 1 1 The Reserved SB is 2 A failure occurs on SB 0 allocated to SB 1 The Reserved SB is incorporated through a reboot 3 Partition 1 enters the stopped state IOB I O board GSPB Giga LAN SAS and PCI_Box Interface Board Figure Example of Reserved SB operations 31 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Revision History Edition 2010 03 31 Revised location 1 Description 2010 08 10 5 Cluster Configurations in a Virtual Environment Addition of a chapter 2011 04 28 All pages Support of new models 2011 06 30 All pages Support of RHEL6 and VMware 1 Chapter section and item numbers in the Revised location column refer to those in the latest edition of the document 32 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Terms of Use Copyrights trademark rights and other intellectual property rights The content including text images and audio is protected by copyrights trademark rights and other intellectual property rights The content can be printed or downloaded for personal use only Fujitsu or the rights holder reserves the right to2. Cascade am as us os O Disk array unit Disk array unit Figure 3 1 Example of Fibre Channel connections to external storage systems 18 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Notices Please note the following regarding external storage systems in cluster system configuration e For external storage system connection limits and available protocols see the relevant documents e If the ports of all switches are insufficient to keep server storage connections redundant use cascade connections For details of relevant switch specifications see documents for the relevant switches e Fujitsu recommends using ETERNUS multipath driver MPD for ETERNUS connections PRIMECLIUSTER can support MPD paths in a range of two to eight e Notes on immediate cluster switching on I O unit failure It may take time to complete the fail over operation if an I O unit containing a failed system disk If an ETERNUS or Fibre Channel fails and ext3 file system is used with the default setting PRIMECLUSTER tries as long as possible to continue system operation Since system operation is based on information kept in memory the system may not immediately switch to standby mode To switch clusters immediately on failure of an I O unit containing system volumes set the parameters below Specify errors panic for the mount option of ext3 on each part
3. User SVmco User Manual PRIMEQUEST 1000 Series General Description General Description PRIMEQUEST 1000 Series Installation Manual Installation Manual PRIMEQUEST 1000 Series Administration Manual Administration Manual Trademarks Linux is a trademark or registered trademark of Linus Torvalds in the United States and other countries Red Hat and all Red Hat based trademarks and logos are trademarks or registered trademarks of Red Hat Inc in the United States and other countries VMware is a trademark or registered trademark of VMware Inc in the United States and other countries Intel is a trademark or registered trademark of Intel Corporation or its subsidiaries in the United States and other countries Oracle and Java are registered trademarks of Oracle Corporation and or its subsidiaries or affiliates in the United States and other countries NetVault is a registered trademark of BakBone Software Inc in the United States and other countries Other company names and product names are the trademarks or registered trademarks of their respective owners Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Contents PY OTA CE AE E EEE E E centred ta aeeravtnls 1 Ab rev a TIONS anaa A a a a ea aa Ca cbse aa Ee eaa O a aae eai 1 Related DOCUMEN Skein ereitea eiea a Eae a E a dapa 2 TradematkS aeni neie e oiis a pi a EE E EEES EE S E Es 2 1 Configuration of Cluster S
4. components installed in hardware partitions v MMB Management LAN MMB Management LAN monitors all the hardware components in the server e Production LAN Production LAN is used by applications to exchange data e Cluster interconnect Cluster interconnect is used by cluster components in each node to monitor each other s node states Cluster interconnect must be separate from the production LAN Use separate segments for the management LAN production LAN and cluster interconnect LAN Production LAN Cluster interconnect Cabinet Partition Partition MMB l User REMCS CE z c l a a a a aal Maintenance LAN Management server Figure 2 1 Network types Cabinet MMB 0 User REMCS cE Port Port Port I i MMB 5 User REMCS cE ea Port Rot Part REMCS CE Port Port Partition management LAN erreren MMB management LAN REMCS center 12 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 2 2 Cluster Operations of Management LAN and Partition Management LAN Management LAN is used for monitoring and operating the cluster system by inter working with PRIMEQUEST hardware and PRIMECLUSTER Node state monitoring If a system panics or is powered off or if other problem occurs without warning this function can help detect such problems immediately Forcible shutdown T
5. not make all duplicated PCI cards inaccessible Always ensure I O components related to PCI slots are fully separated e RAID level DO NOT use RAIDO for external storage system For details of RAID levels refer to the documents for each respective storage system e Recommendations for redundant controllers and paths Fujitsu recommends that each storage system consists of dual controllers and redundant paths e To avoid performance degradation due to access conflicts on shared storage system take the following measures Do not use too many storage systems to reduce performance overheads y Deploy different databases or files to different disks for improved parallelism DO NOT allot too many disk drives to one interface to avoid overloading the disk interface 3 2 Tape Device Fibre Channel is usable for sharing an LTO library unit between multiple nodes Backup software is required when using a tape device or library unit For details on backup software see 4 Backup Restore 17 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 3 3 Connections to Shared I O Units 3 3 1 External Storage System connections PCI cards and cluster system paths should be redundant 1 Duplicate PCI cards 2 Duplicate paths between nodes and switches and between storage systems and switches 3 Duplicate controllers of external storage systems 5 5 E FC SWITCH Hy LA
6. 4 Red Hat Enterprise Linux 5 for x86 Red Hat Enterprise Linux 6 for Intel64 Red Hat Enterprise Linux 6 for x86 VMware vSphere 4 System Board Linux RHEL Management Board PRIMEQUEST Server Agent ServerView Mission Critical Option Network Interface Card Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Related Documents See the following manuals as needed when designing the cluster configuration Name Abbreviation PRIMECLUSTER PRIMECLUSTER Installation and Administration Guide Installation and Administration Guide PRIMECLUSTER Concepts Guide Oracle Solaris Linux PRIMECLUSTER Cluster Foundation CF Linux Configuration and Administration Guide PRIMECLUSTER RMS with Wizard Tools Oracle Solaris Linux Configuration and Administration Guide PRIMECLUSTER RMS Oracle Solaris Linux Reference Guide PRIMECLUSTER Web Based Admin View Operation Guide PRIMECLUSTER GLS Configuration and Administration Guide PRIMECLUSTER GDS Configuration and Administration Guide PRIMECLUSTER GDS Configuration and Administration Guide PRIMECLUSTER GFS Configuration and Administration Guide PRIMECLUSTER Wizard for Oracle Configuration and Administration Guide Sr 1000 Series ServerView Mission Critical Option
7. ED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 6 2 Notes on PRIMECLUSTER PRIMEQUEST 1000 series supports PRIMECLUSTER 4 3A00 and later For PRIMECLUSTER with Oracle Database see the following manual for investigating availability of operating systems and Oracle Database PRIMECLUSTER Wizard for Oracle Configuration and Administration Guide When configuring cluster set slew mode for NTP to synchronize NTP time In the cluster system environment Fujitsu recommends connecting two UPS units to each server and each external storage equipment If UPS is not properly connected to equipments above failover of clusters may not work properly in the event of power shortage or power supply failure Physical cluster configurations inter working with hardware mechanisms can detect node error immediately However a virtual environment without such inter working with hardware cannot enjoy such immediate error detection on Guest Operating System 30 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Appendix Reserved SB function This function automatically disconnects a failed SB and restarts the relevant partition with Reserved SB incorporated The Reserved SB is a spare SB which is replaced in the event of a SB failure By automatically incorporating the Reserved SB at partition restart the function prevents performance degradation and enables
8. Gad Seas EESE i E ais aa EES AHO 31 Revision TISIONY iscissi ssecvinissepruisoscassniessanidesosors besope sigoot assa iiea s veseri kesse Kissat 32 TOPINS OF SO AAO EE EEEE EE EE TE 33 4 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 1 Configuration of Cluster Systems for the PRIMEQUEST 1000 Series This chapter describes configuration of cluster systems for the PRIMEQUEST 1000 series 1 1 Reliability Improvement with Cluster System Cluster system helps maximize system operational time by server duplication and swift fail over of application environments on failure It allows continuing system operation even after software failure including operating system middleware and applications by switching all operations on a failed active node to the standby node Such switch over operation is exemplified below C gt Error Active Node A ae fails Node A Node B Node A Node B wn Ps is Q o lt Figure 1 1 Example of two nodes in active standby mode Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER The following table compares operational continuity between a cluster system and single server incorporating a Reserved SB in the event of failure Table 1 1 Comparison between a cluster system and single PRIMEQUEST with Reserved SB Item Cluster system System that incorporates Re
9. IMECLUSTER version4 3 below are usable with PRIMEQUEST Function Product name PRIMECLUSTER Enterprise Edition PRIMECLUSTER HA Server PRIMECLUSTER GDS Table 1 2 PRIMECLUSTER products and functions provided with PRIMECLUSTER PRIMECLUSTER GLS Parallel database Y N N Cluster control HA switching type cluster Software RAID mechanism Y Y Storage control Volume management function Shared file system function Transmission line duplication function Network control Y Supported N Not supported For details on PRIMECLUSTER visit the URL below http www fujitsu com global services computing server primequest products software pcl lineup html These products are outlined below e PRIMECLUSTER Enterprise Edition As a PRIMECLUSTER base software product PRIMECLUSTER Enterprise Edition has various clustering functions such as parallel database and HA functions This product also includes a volume management function called Global Disk Service GDS a shared file system function called Global File System GFS and a network redundancy function called Global Link Service GLS e PRIMECLUSTER HA Server This provides the most common High Availability HA environments for general operational service and applications Multi directional failover and workload management ensures operational continuity In addition it provides for cluster enabled storage man
10. LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Partition X Partition Y Cabinet P Tam s come o camm o cog i FC switch i Cluster interconnect Figure 2 5 Network configuration example for two nodes Fibre Channel connections for an in box cluster Production LAN sses Management LAN s FC cable Notices about network connections Please note the following when designing cluster networks e DO NOT use Management LAN for cluster interconnects e Set up cluster interconnects and keep them separate from the production LAN e DONOT use STP Spanning Tree Protocol in the network switch for cluster interconnects STP is prohibited for network switches because such STP link related reconfiguration may interrupt transmission of cluster interconnect e All components in cluster interconnects must be redundant 16 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 3 Shared I O Unit 3 1 External Storage System Fibre Channel and iSCSI are usable for connection to external storage system Notices e Duplication of connections with external storage system Fujitsu recommends duplicating PCI cards for external storage system connection Fujitsu also recommends that each duplicated PCI card is mounted in a PCI slot so that any single I O component failure PCIe switch chips or wires does
11. PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Preface This manual describes the network and shared I O unit information and configuration required for design of a cluster system using PRIMECLUSTER in Red Hat Enterprise Linux 5 and 6 environments for the PRIMEQUEST 1000 series The PRIMEQUEST 1000 series is a lineup of Mission Critical IA Servers with QPI Quick Path Interconnect network and memory mirror functions Other PRIMEQUEST functions include the partition function and Reserved SB function with which a highly reliable system can be configured For basic information about the PRIMEQUEST 1000 series see the PRIMEQUEST I800E 1800E2 Design Guide For details on the Reserved SB function see Reserved SB function in Appendix Operation with the cluster system for the PRIMEQUEST 1000 series can deliver higher availability by supporting troubleshooting of operating system middleware and application failures Abbreviations This manual uses the following product name abbreviations Product name or component name Abbreviation PRIMEQUEST 1400S PRIMEQUEST 1400E PRIMEQUEST 1400L PRIMEQUEST 1800E PRIMEQUEST 1800L PRIMEQUEST 1000 series PRIMEQUEST 140082 or PRIMEQUEST PRIMEQUEST 1400E2 PRIMEQUEST 1400L2 PRIMEQUEST 1800E2 PRIMEQUEST 1800L2 Red Hat Enterprise Linux 5 for Intel6
12. agement and an HA network This product includes GDS GFS and GLS e PRIMECLUSTER GDS This volume management software protects customer assets from disk failures and other unexpected events by mirroring the system and data storage disk devices e PRIMECLUSTER GDS Snapshot This optional software can greatly reduce operational stoppage time during the disk backup restore process by 10 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER instantly creating a snapshot of a disk or volume e PRIMECLUSTER GLS This software product offers network redundancy by multiplex NICs network interface cards GLS is the optimal for a system with high network continuity 11 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 2 Networks 2 1 Network Types This section describes the network required for administrating PRIMEQUEST 1000 series For the network settings of PRIMEQUEST 1000 series see the Administration Manual e Management LAN Management LAN is used for monitoring and operating target servers and software products on those servers via a management server which collects management data from hardware partitions and MMB in the target servers v Partition Management LAN Partition Management LAN monitors the status of operating systems applications and middleware in hardware partitions It can also monitor cluster
13. approve or withhold permission for use for any other purpose e g for reuse on your own webpage or upload to another server Limitations of warranty Fujitsu provides no guarantee of the accuracy merchantability or fitness for use of the content Fujitsu assumes no legal responsibility for any damage arising from use of the content The content in this manual may be revised or deleted without prior notice For clarification of any unclear points send us your comments or inquiries by e mail to primeweb primequest ml css fujitsu com All rights reserved C122 A007 04EN June 2011 33 Copyright 2012 FUJITSU LIMITED
14. be used with PRIMECLUSTER GDS or PRIMECLUSTER GDS Snapshot for backup purposes e To save backup data collected from a guest operating system to tape or other external storage media transfer the backup data to hypervisor and then save the data from hypervisor to the external storage device 21 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 4 2 Data Volume Backup Restoration You can back up user data during cluster operation by using dedicated backup software The following figure shows a configuration example LAN Active system Standby system Partition n Client Agent software Backup server PRIMERGY Server Manager software ETERNUS DXxx Snapshot gt Shared disk Master Shared disk Proxy Figure 4 1 Backup configuration example Under control of Server Manager software Client Agent software in a management node backs up data on shared disks to the backup device PRIMECLUSTER GDS Snapshot can also be used to collect backup data from one disk master to another disk proxy in an ETERNUS storage system Supported tape devices and other equipment vary depending on software products used For details see the corresponding product manuals and handbooks 22 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER The software products liste
15. continuity of virtualized configuration patterns in the event of failure Table 5 1 Comparison of redundancy in the KVM environment Failure occurrence location Configuration pattern One Management Multiple Management Multiple Management OS OS in box cluster OS multi node cluster Hardware component outside all hardware partitions e g I O unit Hardware component contained in hardware partition e g SB HDD Management OS Guest OS Y Operation continues by automatically switching to the standby system N Operation stops 1 PRIMECLUSTER 4 3A10 and later provide a switching function in the event of error in Management OS If this function is not used only manual switching is available Table 5 2 Comparison of redundancy in the VMware environment Failure occurrence location Configuration pattern Multiple Management OS in box cluster Multiple management OS multi node cluster Hardware component outside all hardware partitions e g I O unit Y Hardware component contained in hardware partition e g SB HDD ESX host Guest OS Y Operation continues by automatically switching to the standby system N Operation stops 28 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 6 Note 6 1 Notes on PRIMEQUEST 1000 Series For in box cluster configuratio
16. d below or a network storage server can be used for such purposes The products below can help backup operations continue painlessly even after a failover occurs e PRIMECLUSTER GDS Snapshot e ETERNUS SF AdvancedCopy Manager In a cluster where ETERNUS SF Advanced Copy Manager is installed on the production server and other backup software is installed on the backup server ETERNUS SF Advanced Copy Manager creates replicas copied data which can be backed up directly from the backup server Therefore no special attention needs to be paid to the cluster 23 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 5 Cluster Configurations in Virtual Environments 5 1 Types of Cluster Configurations Using Virtual Machine Functions Using PRIMECLUSTER virtualized servers can be clustered in the same way physical servers PRIMECLUSTER supports the following virtual environments e KVM for Red Hat Enterprise Linux 6 e VMware Cluster configurations for virtual environments are as follows e Configuration with one Management Operating System See 5 1 1 Configuration with one management operating system e Inbox cluster configuration with multiple Management Operating Systems See 5 1 2 Configuration with multiple management operating systems e Multi node Cluster configuration See 5 1 2 Configuration with multiple management operating systems For detail
17. e Channel connections with a tape device 20 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 4 Backup Restoration 4 1 System Volume Backup Restoration In a local boot configuration PRIMECLUSTER GDS and PRIMECLUSTER GDS Snapshot help you back up system volumes at any time during system operation For details of PRIMECLUSTER GDS and PRIMECLUSTER GDS Snapshot see PRIMECLUSTER Global Disk Services Description Also be sure to create backup data before installation upgrade or removal of software products and before modification of software or hardware settings If interruption of a node is allowed for backup you can store such backup data on an external storage system However you need to notice the node stoppage and follow a predetermined operational procedure for the clustered system Also for restoration of backup data follow the general restoration procedure Backup data must be created each time you modified the cluster configuration If data taken before such a modification is mistakenly restored an error occurs during the restoration process The cluster also stores information specific to each node This means nodes may store different information Make sure that backup and restoration operations are executed for all nodes For the details of cluster system setup see the relevant PRIMECLUSTER manuals e Note only Red Hat Enterprise Linux 6 for Intel 64 can
18. e cluster in Cabinet 0 incorporates the Reserved SB Cabinet 1 switches from and is recovered as a standby standby to active system saa Partition a z E Cluster configuration IOB I O board GSPB Giga LAN SAS and PCI_Box Interface Board Figure 1 2 Incorporating a Reserved SB in cluster configuration Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 1 2 Cluster Configurations PRIMEQUEST 1000 series has two types of cluster system One is a multi node cluster where the cluster system is configured with multiple servers The other is an in box cluster where the cluster system is configured using multiple partitions in a single server For details of cluster system configurations in a virtual environment see 5 Cluster Configurations in a Virtual Environment 1 2 1 Multi Nodes Cluster Multiple node clusters provide full redundancy by unifying multiple servers into the cluster system Plus it maintains system operation while maintenance work takes place as one server can be configured out of the system for repair This type of configuration has higher redundancy and reliability than in box cluster configurations because of its full redundancy Active system Standby system Node 0 i Figure 1 3 Multi node cluster Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 1 2 2 In Box Clus
19. his function can prevent user resource contention by intentionally triggering a panic in a problematic node and resetting it The Partition Management LAN is connected to a console with Web Based Admin View which is used for operational management for the cluster system For details on Web Based Admin View visit the URL below using search string PRIMECLUSTER and view the PRIMECLUSTER Web Based Admin View Operation Guide for PRIMECLUSTER 4 3A00 or later http www fujitsu com global services computing server primequest documents pcl manuals html 2 3 Cluster Interconnect PRIMECLUSTER can recover operations of failed nodes swiftly by finding the problem and transferring all information required for fail over Cluster interconnect is used to monitor the status of all nodes to find problems at the earliest possible moment and for transfer of information for fail over 2 3 1 Network Equipment Cluster interconnect configuration example Fujitsu recommends cluster interconnect paths be fully redundant In other words any failure in one interconnect path must not affect any other interconnect path Otherwise component failure can make all interconnects inaccessible For instance two PCI slots linked to a common PCI Express switch will become unusable if the PCI Express switch fails For full redundancy I O components linked to PCI slots should be fully separated from each other Fujitsu recommends one interconnect be drawn from GSPB and a
20. ition contained in the system volume As a result as soon as the ext3 file system is notified of the I O error caused by the I O unit failure it will trigger an operating system panic and switch clusters However a certain amount of time may still be needed to notify the ext3 file system of the I O error Note that clusters are not switched immediately in that case Periodic writing to system volumes helps detect I O errors early time Example of etc fstab specification directories such as var and home are contained in one system volume LABEL ext3 LABEL boot boot ext3 LABEL var var ext3 LABEL home home ext3 19 errors panic errors panic errors panic p en p pd A U N me errors panic Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 3 3 2 Connections with Tape Device Maximum four nodes for two active standby nodes and cascade connections can share a tape device e Only one node can access a tape device at a time e One network switch dedicated for connection with the tape device is required e A PCI card dedicated for tape device connections is required DO NOT share it with other file connections The following figure explains Fibre Channel connections PAR E L ph FC F L ERME ETE EEE FC SWITCH nE O q a oO Cascade Zoning m Disk array unit Tape device Figure 3 2 Example of Fibr
21. nother drawn from NIC mounted in the PCI slots below e IOB PCI slot or e PCI_Box PCI slot All nodes remain connected even if any I O component such as a GSPB IOB or PCI_Box fails 13 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 2 3 2 Cluster Interconnect Configuration Cluster interconnect multiplexing Multiplexing interconnects can improve system availability The following figures exemplifies interconnect Figure 2 2 Example of the interconnect of two nodes connections Hub or switch Hub or switch Figure 2 3 Example of the interconnect of three nodes A hub or switch should be used to connect three or more nodes 14 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 2 4 Network Configuration Examples Figures below explain cluster system topology for multi node and in box clusters In these figures LAN networks are configured fully redundantly This means cluster interconnects use different I O components one LAN port on GSPB and another NIC mounted on IOB Cabinet Cabinet FC switch Cluster interconnect Production LAN snneeenenees Management LAN FC cable Figure 2 4 Network configuration example for two nodes Fibre Channel connections for a Multi Node Cluster 15 Copyright 2012 FUJITSU
22. ns of PRIMEQUEST 1800E and 1800E2 Fujitsu recommends that components such as IOB and GSPB are not shared between hardware partitions Such separation of IOB and GSPB greatly improves operational continuity of the clustered system PSA or SVmco must be installed in each hardware partition e PSA or SVmco is required for operation of PRIMEQUEST in all cluster system nodes Cluster switching fails if PSA or SVmco is not installed properly e PSA is available with PRIMEQUEST 1800E SVmco for PRIMEQUEST 1800E2 With single MMB every MMB failure triggers fail over However the fail over is suspended and the entire system remains interrupted until administrator operates fail over So replacement of faulty a MMB is necessary for immediate and automatic fail over When configuring a cluster be sure to set the parameters below e Physical IP address settings of MMB MMB 0 and MMB 1 See the following document 3 3 6 Configuring the MMB Network in the Installation Manual e IP address setting of Management LAN See the following documents according to the model used y PRIMEQUEST 1800E 6 2 7 Setting Management LAN IP address in the Installation Manual PRIMEQUEST 1800E2 SVmco User Manual e Firewall function settings See the following documents according to the model used y PRIMEQUEST 1800E 6 2 4 Checking the firewall function opening ports in the Installation Manual PRIMEQUEST 1800E2 SVmco User Manual 29 Copyright 2012 FUJITSU LIMIT
23. s and notes on cluster systems using the virtual machine function see the PRIMECLUSTER Installation Guide e For the KVM environment translate management operating system as hypervisor e For the VMware environment translate management operating system as ESX host Notices In a virtual environment multiple cluster systems can share a physical NIC Virtual bridges need to be created separately 24 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 5 1 1 Configuration with One Management Operating System Fujitsu recommends this configuration is only used for test environments One physical server serves as the cluster system where applications can continue operation in the event of a failure in a Guest Operating System e This configuration cannot be used in VMware environments e The configuration requires virtual network entity definitions for inter connecting Guest Operating Systems note it is designed for test environment use PRIMEQUEST cabinet Partition n PRIMEQUEST virtual machine function Virtual machine Virtual machine Guest domain Guest domain Application ae memory set icp nik a Management LAN Penans cater i Production LAN S raaa Figure 5 1 Example of a configuration with one Management Operating System 25 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration De
24. seeeceeseaeeceeseseeeeesseaeeees 18 TOES a5 aye E Ra ary et sh Ra aN Cae T Soba cece ease a A ge ae St 18 3 3 1 Connections with External Storage System ssssssssssssssssssssssssssssssssssee 18 3 3 2 Connections with Tape Device cccccecccccccesssseeeececcceeesssssescececeeeusssesescees 20 4 BACKUP IRESLOFALON desiree daRarUii wine rincAormenrnkeird 21 4 1 System Volume Backup Restoration sssesssesssesssesssssssssssssssesssesssssssesssesssessse 21 4 2 Data Volume Backup Restoration sensein ara A a 22 5 Cluster Configurations in Virtual Environimentt ccssccsesssssscesscescesssssssseseees 24 5 1 Types of Cluster Configurations Using Virtual Machine Functions 24 5 1 1 Configuration with One Management Operating System 0eee 25 5 1 2 Configuration with Multiple Management Operating Systems 26 5 25 Reliability Comparison issih ccc cck on Season eh SOA SESS 28 3 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 6 CIN OUCOS cessione tienaan enera senyatunntowkcnssantatsvacheaterenausdasuianvaessiaGoinabiaueins ERa eisiata 29 6 1 Notices on the PRIMEQUEST 1000 Series ssssssssesesssssssesssssssssssssssesssssssessse 29 6 2 Notices on PRIM EC LUSTER css vavehjecenutclait acenechea say lena i vesbins lub sunentacetutelaates eens 30 PLP CNIS 5 ccivass isncnssusninsssvsvanasaussan nesei i ER kein daa
25. served SB Hardware CPU failure Memory Can continue operation Need to reboot system 5 OS Software Middleware failure Application Can continue operation Need to stop system ie Summary Even if hardware or software fails operation can continue by switching to the standby system 1 Operation continues by switching to the standby system 2 Operation continues after a reboot of the relevant partition 3 Operation stops After an SB failure operation can resume quickly Recovery can be done promptly when a Reserved SB is used for cluster configuration Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER Reference Prompt recovery of the standby node If any SB CPU or memory fails recovery time can be greatly reduced by the incorporation of a Reserved SB in cluster configuration If a SB on active node fails the standby node is activated so operation can be resumed immediately The failed node is changed to standby mode in parallel The Reserved SB can help the node on the standby mode can recover promptly because the Reserved SB can automatically disconnect the faulty SB and incorporates a spared SB Active Failure Stopped Reserved SB Reserved SB Cabinet 0 LospB_oaf LGSPB_0B 1 Cabinet 0 and Cabinet 1 2 A failure occurs on an SB 3 Through a reboot Cabinet 0 configure th
26. sign Guide Linux PRIMECLUSTER 5 1 2 Configuration with Multiple Management Operating Systems In this configuration applications can continue operation even if a hardware network or disk failure occurs e In box Cluster PRIMEQUEST cabinet Partition m PRIMEQUEST virtual machine function Partition n PRIMEQUEST virtual machine function Virtual machine Virtual machine Guest domain Guest domain Application ee aes Production LAN Se a ee ee en Figure 5 2 Example of a configuration with two Management Operating Systems in box cluster 26 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER eMulti nodes Cluster PRIMEQUEST cabinet PRIMEQUEST cabinet Partition n Partition m PRIMEQUEST PRIMEQUEST machine function virtual machine function Virtual machine Virtual machine Guest domain Guest domain wa eww ecco weno sseseees ame e econ see see seceeed eee eee eee eee ne eee ee eee eg eee eee eny Figure 5 3 Example of a configuration with two Management Operating Systems by multi node cluster 27 Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 5 2 Reliability Comparison Management OS and hardware virtualized environment redundancy greatly affects operational continuity The tables below compare operational
27. ter In box clusters can much improve redundancy by clustering multiple partitions inside a single server Although the costs for system creation and maintenance can be reduced a failure of a non redundant component I O unit may cause the whole system to stop ir Active system Node 0 Standby system Node 1 Figure 1 4 In box cluster 1 2 3 Supported Configurations Hardware configuration e All nodes in the cluster system should have the same configuration including number of CPUs CPU frequency memory capacity hard disk capacity and number of hard disks e Server models in the cluster system must also be the same e PRIMEQUEST 1800E and 1800E2 can support both multi node and in box clusters e The number of required nodes varies depending on the cluster operation mode For details on cluster operation modes see the PRIMECLUSTER Installation Guide Software configuration Please note server management software varies depending on the model Table 1 2 Supported configuration for each model software PRIMEQUEST Server management software Model 1800E PSA 1800E2 SVmco Copyright 2012 FUJITSU LIMITED PRIMEQUEST 1000 Series Cluster Configuration Design Guide Linux PRIMECLUSTER 1 3 PRIMECLUSTER PRIMECLUSTER and PRIMEQUEST in combination provide high availability such that a hardware partition or server in stand by mode can take over operation in the event of a failure Functions of PR
28. ystems for the PRIMEQUEST 1000 Series 5 1 1 Reliability Improvement with Cluster System ccccccessessesseseseseseseseeeeeeeees 5 1 2 Cluster Configurations ccccccccesssessseeseeseeeeeeeeeceeeeeeeeeeeeeseeeeeeeeeeeseseeeeeeeeeeeees 8 1 2 1 Multi Nodes Cluster arrisera iei eaea E aa inda 8 1 2 2 Tn Box Clusters senean iei e T a E E E toraeedetades 9 1 2 3 Supported Configurations ccccccccesssesseeessessseeseeeceeeseeeeeseeseeeeeeeeeeeeseeeeeees 9 13 PRIMECLUS TER aieeaa ane aaee a A a T A E AE AER OAA AE TEREE eiperee Seamed as 10 De SIN CLW A AEREE E E E TEE E E A E EEE EEE AT 12 2 1 Network Ty pesiccis ciscccscs sissies esecoc haa esa she Bea eA e oo diaa 12 2 2 Cluster Operations of Management LAN and Partition Management LAN 13 2 3 7 Cl ster Interconnected entie ae es cosh ch ends ite sae ak ua da be oas dessa gb eae Ta a 13 2 3 1 Network Equipment Aa A A A AE Ae a E IRER 13 2 3 2 Cluster Interconnect Configuration sssssssssssssssssssssssesssesssssssesssssssesseo 14 2 4 Network Configuration Examples ssssessssesssesssesssssssssssesssesssssssssssesssesssessne 15 3 Shared T O Units spiesen tensona esatea siias eats eaan ee 17 3 1 External Storage System cccccccsscceceeecceeceeececeseeeseeeseeeseeeseeaseesseeseeeaseeaaeees 17 Od Pape WEVa Cette cS tical A A A S EEE AE Satria ncdneds moans E A A Ss 17 Connections with Shared V O Unit eeeeceecsssseecesssceecee
Download Pdf Manuals
Related Search